The Challenge of Scientific Open Data

The issue with scientific open data is not just about making it available. It is also about being able to find useful datasets, understanding their structure, verifying their licenses, and checking the contents of the files before starting an analysis.

To address this, AWS has made available the RODA MCP Server, a new open-source MCP server that connects AI assistants to the Registry of Open Data on AWS.

The solution exposes a series of tools through the Model Context Protocol that compatible assistants can invoke during a conversation. AWS explicitly mentions Kiro and Claude Code, but the server can also be connected to other MCP clients.

Search, Preview, and Sampling of Datasets

Once configured, the server allows searching for datasets by topic, organization, license, or keywords. It can return metadata, access methods, and related resources, as well as identify similar repositories based on shared tags.

For completely public datasets, it can even show the structure of Amazon S3 buckets and read a sample of the files directly within the conversation. The current implementation limits sampling to the first 100 KB of the file: an amount sufficient to check format, variables, and data organization, but not enough for actual analysis.

A researcher can therefore ask what genomic datasets are available under a certain license, search for satellite data on soil temperature, or view the structure of the 1000 Genomes project without having to manually sift through catalogs, documentation, and buckets. AWS claims that the system can reduce the time needed to locate data from hours to just a few seconds. However, the company has not published benchmarks or independent measurements. The verifiable advantage mainly lies in consolidating, within a single conversational interface, tasks that normally require several steps.

Over 1,100 Datasets with Different Access Methods

As of June 2026, the Registry of Open Data on AWS had reached 1,122 datasets, amounting to over 400 petabytes of data made available by more than 400 organizations. These providers include NASA, NOAA, National Institutes of Health, and Allen Institute. The datasets are hosted on AWS infrastructure but remain the property and responsibility of the organizations that publish them.

The definition of “open data” does not imply that every repository is accessible unconditionally. The repository distinguishes three categories:

• Public datasets that can be accessed without an AWS account;
• Datasets that require credentials and may involve paying transfer or computation costs by the user;
• Controlled access archives, especially in healthcare, for which further permissions are necessary.

Preview and sampling functions are only available for the first category. Licenses must still be verified individually before using the data.

Code on GitHub with Apache 2.0 License

The RODA MCP Server is published in the AWS Labs MCP Server repository under the Apache 2.0 license. Installation requires Python 3.10 or higher and the uv package manager. AWS specifies that the software is intended for development, testing, and evaluation activities and is provided without guarantees regarding the quality or reliability of the results. The repository also notes that language models can make mistakes and that responses should be verified before using the system in operational environments.

"Open data represents the foundation of scientific progress, but finding the right dataset shouldn't require specialized technical skills. This tool puts artificial intelligence at the service of researchers worldwide, whether they are dealing with climate change, mapping disease spread, or sequencing genomes, allowing them to spend less time searching for data and more time making discoveries for the benefit of all," comments Werner Vogels, CTO of Amazon.

"I am truly excited that the new AWS RODA MCP Server is publicly available. We were able to quickly integrate it into the Queryable Earth platform by Element 84, and now we can deploy agents capable of locating, previewing, and processing data hosted in the AWS Open Data repositories. Open data, open specifications, and GeoAI allow us to completely rethink how we extract useful information for emergency management and support decision-makers," states Dan Pilone from Element 84.

"The greatest opportunity of AI applied to science arises from the intersection of organizations, datasets, and disciplines, where unexpected connections can lead to entirely new discoveries. Tools like the RODA MCP Server foster this 'serendipity,' providing scientists and their AI agents seamless and immediate access to a wide variety of reliable data, enabling rapid interdisciplinary innovation," explains David Feng from the Allen Institute.